When to Use
- To differentiate multiple sclrosis (MS) from other conditions
-
MRI data needed
- Brain Lesions assessment using T2w-FLAIR and T1w images
- (Optional) Perivenular lesion assessment using T2*w-EPI magnitude and FLAIR images
- (Optional) Cortical lesion assessment using T2w-DIR, T1w-PSIR, T2w-FLAIR, and/or T1w-MPRAGE images
- (Optional) Paramagnetic rim lesions assessment T2*w-EPI magnitude and phase
- This tool is for research purposes only and is not approved by the FDA or EMA. The authors assume no responsibility if used by clinicians as a primary diagnostic tool.
- More details can be found in Borrelli et al. Neurol Neuroimmunol Neuroinflamm. 2024
Pitfalls
All machine learning models were trained on a medium-sized (n=285) multi-centric dataset of MS and MS-mimics patients.
-
Specific MRI protocol
- CVS and PRL were assessed using sub-millimetric 3D T2*-EPI
- CL were assessed using 3D-DIR and 3D-MPRAGE for centres for which DIR was available, and synthetic-DIR (generated from T2/T1 images) and 3D-MPRAGE when DIR was not available
- Dataset includes adult MS population with a large variety of phenotype and disease duration.
Why Use
Our results indicate that using advanced MRI biomarkers significantly improves MS diagnostic balanced accuracy, increasing specificity while maintaining the high sensitivity of current McDonald DIS criteria.
- CVS emerges as the most diagnostically powerful biomarker, with PRL and CL closely behind
- Models combining CVS, CL, and PRL show the highest MS diagnostic performance
- Simplified more practical biomarker assessments were competitive, showing no significant difference with full-count assessment
- The best model uses %CVS, #CL and #PRL as input
- The best model using simplified assessment takes Select3*, CL1 and PRL1 as input
- More details in the upcoming paper.